The CDC’s Center for Forecasting and Outbreak Analytics has been actively deploying AI-enhanced surveillance and modeling tools throughout 2026 to assess emerging infectious disease risks, including evaluating the danger a 2026 Ebola outbreak caused by the Bundibugyo virus in the Democratic Republic of the Congo poses to the United States, and separately assessing a 2026 Andes virus outbreak that occurred aboard a cruise ship. The center, established in the wake of the COVID-19 pandemic’s forecasting failures, represents the federal government’s most concentrated bet that predictive analytics and machine learning can shorten the lag between an outbreak’s earliest signals and a coordinated public health response.
The underlying science supporting this approach has produced measurable results. A peer-reviewed study integrating machine learning-based diagnostic predictions with traditional epidemiological surveillance, using 4.5 million patient records collected between 2010 and 2022, found that among diseases that experienced five or more outbreaks during that period, AI-enhanced surveillance detected 33.3% of those outbreaks earlier than conventional methods — 41 out of 123 tracked outbreaks — with lead times ranging from one to 24 days.
Why a Few Days of Lead Time Actually Matters
In outbreak response, the difference between detecting a cluster of infections on day one versus day ten can determine whether public health officials can still contain spread through targeted measures like contact tracing and localized interventions, or whether the outbreak has already grown too large for anything short of broader restrictions. The 24-day maximum lead time reported in the biosurveillance study, while modest-sounding, represents a meaningful head start in outbreaks where case counts can double every few days.
How the AI Models Actually Work
Rather than waiting for confirmed lab-diagnosed cases to accumulate — the traditional trigger for outbreak recognition — these systems scan electronic health records for patterns of diagnostic codes and symptom presentations that historically preceded confirmed outbreaks of a given disease, flagging unusual clusters before formal diagnostic confirmation catches up. This approach, sometimes called syndromic surveillance when applied to symptom data, has been supercharged by machine learning’s ability to detect subtle statistical anomalies across millions of records that a human epidemiologist reviewing spreadsheets would likely miss.
Why This Center Exists at All
The Center for Forecasting and Outbreak Analytics was created specifically in response to widely documented failures in the United States’ ability to model and forecast COVID-19’s spread in near real time, failures that public health leaders acknowledged left officials making major policy decisions with outdated or incomplete data during the pandemic’s most critical early months. Its mandate combines traditional epidemiological modeling with newer machine learning approaches specifically because agency leadership concluded that the older, slower forecasting methods used going into 2020 were not adequate for the speed at which a modern outbreak, amplified by global travel and dense urban populations, can actually move.
The Two Live Test Cases
The Bundibugyo virus Ebola outbreak in the DRC represents a classic cross-border risk assessment scenario the center is built for: evaluating how likely international travel could import cases into the US healthcare system, and what surveillance posture is warranted given the outbreak’s trajectory in its country of origin. The Andes virus cruise ship cluster is a different kind of test — a contained, enclosed population where an outbreak’s spread dynamics differ sharply from a community-based epidemic, requiring models tuned to that specific transmission environment rather than general population outbreak patterns.
The Skepticism Public Health Officials Still Carry
Epidemiologists caution that AI-enhanced biosurveillance, while promising in retrospective studies, has not yet been tested at scale against a genuinely novel pathogen the way it eventually will be — most of the validating research to date analyzes known disease patterns in historical data, which is a different and easier problem than detecting something entirely new. There’s also a persistent tension in outbreak forecasting between sensitivity and specificity: a system tuned to catch outbreaks earlier will also generate more false alarms, and public health agencies have limited political and resource capacity to chase down alerts that don’t pan out.
What This Means Going Forward
The center’s active use of these tools on two simultaneous, geographically distinct threats in 2026 suggests AI-assisted biosurveillance has moved from pilot demonstration into the CDC’s operational toolkit, rather than remaining a research exercise. Whether that translates into faster real-world containment will likely only become clear in the response to the next outbreak that tests the system’s lead-time advantage against an unfamiliar pathogen, rather than one the models have already learned to recognize.
